Flexible Top Trading Cycles and Chains Mechanism: Maintaining Diversity in Erasmus Student Exchange

1/52 Flexible Top Trading Cycles and Chains Mechanism: Maintaining Diversity in Erasmus Student Exchange Umut Dur Onur Kesten Utku Ünver Econometric Society World Congress'15

New Market-Old Mechanism 2/52

2/52 New Market-Old Mechanism Analyze the student exchange program in Europe

2/52 New Market-Old Mechanism Analyze the student exchange program in Europe Propose a variant of the Top Trading Cycles and Chains mechanism to x the observed problems

3/52 Student Exchange in Europe Student exchange programs enable college students to study one semester at another member college

3/52 Student Exchange in Europe Student exchange programs enable college students to study one semester at another member college Students pay tuition to their home colleges and they do not pay extra tuition fee to the colleges they visit European student exchange program, Erasmus Programme, is the largest exchange program More than 4,000 member colleges from 33 countries in Europe Each year, around 300,000 students benet from it The purpose of Erasmus is to improve the quality of higher education and strengthen its European dimension. It does this by encouraging transnational cooperation between universities. European Comission's Website

How Successful is the Program? Less than 1.5% of the college students benet from the program 8.0% Share of Outgoing Erasmus Students of the Total College Student Population 2010-2011 7.0% 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% BE BG CZ DK DE EE IE EL ES FR IT CY LV LT LU HU MT NL AT PL PT RO SI SK FI SE UK HR IS TR LI NO 4/52

How Successful is the Program? Lack of diversity in exchanges: Western European countries exchanging students between each other Eastern European countries exchanging students between each other 100.0% Share of Exchange Students Coming from Western Europe 2010-2011 90.0% 80.0% 70.0% 60.0% 50.0% 40.0% BE ES FR IE IT LU UK CZ LV LT HU SK LI TR 5/52

6/52 How Successful is the Program? Dierence between the number of incoming & outgoing exchange students 2004/05 2005/06 2006/07 2007/08 2008/09 2009/10 2010/11 Total UK 9052 9,264 9,273 8,452 8,636 8,770 8,927 62,374 Sweden 3,928 4,532 4,827 5,403 5,793 6,060 6,348 36,891 Denmark 2,087 2,684 2,958 3,292 3,625 3,934 4,262 22,842 Poland -6,058-6,911-7,489-7,744-7,256-6,079-4,640-46,177 Germany -5,154-5,959-6,006-5,752-5,685-6,102-6,059-40,717 Turkey -843-2,024-3,117-4,475-4,560-5,117-5,209-25,345

7/52 How Successful is the Program? Although participation in Erasmus and other European programmes would continue, the imbalances in student ows would remain and probably increase, perhaps intensifying the simmering resentment that UK universities are losing more than they are gaining from these programmes. timeshighereducation.co.uk Danish taxpayers should not have to pay for educating exchange students in Denmark. universitypost.dk

8/52 How Successful is the Program? Under the current practice Low participation: Less than 1.5% of the students beneting Lack of diversity in exchanges: Western Europe vs. Eastern Europe Huge imbalance between the number of incoming and outgoing students

9/52 In This Paper In this paper our objectives are: Increase the number of exchanges Increase the level of diversity Maintain a reasonable level of dierence between the # of incoming and outgoing students We introduce TTCC mechanism under various constraints such as College specic exchange quota Student type specic minimum quota To our knowledge, rst application of TTCC in many to one matching markets

10/52 Outline Current Procedure Model Generalized Top Trading Cycles and Chains Minimum Quota

Current Procedure 11/52 How Does It Work? Currently, in Erasmus programme the exchanges are done through 3 step procedure Bilateral Agreement Application Assignment

Current Procedure 12/52 How Does It Work? Bilateral Agreement Each member college signs bilateral agreement with other members Colleges agree on the maximum number of students to be exchanged Bilateral agreements give reserved seats to the colleges at the other ones Application Each student applies to his home college Ranking over the colleges which signed bilateral agreement with his home college Assignment Each college ranks its own applicants according to GPA, written exam, etc... Each college assigns its own students to the reserved seats via Serial Dictatorship mechanism

Current Procedure 13/52 How Does It Work? Bilateral agreement stage is crucial

Current Procedure 13/52 How Does It Work? Bilateral agreement stage is crucial Once a college signs bilateral contracts it loses its control over the number of incoming students

Current Procedure 13/52 How Does It Work? Bilateral agreement stage is crucial Once a college signs bilateral contracts it loses its control over the number of incoming students Moreover, a college cannot force its own students to participate in the program A college may have more incoming students than outgoing and this causes Increase in class size Need more instructors Financial burden: Less tuition for more students

Current Procedure 14/52 What the colleges do to prevent imbalances? They limit the number of students to be exchanged They do not sign bilateral agreement with some colleges

Current Procedure 14/52 What the colleges do to prevent imbalances? They limit the number of students to be exchanged They do not sign bilateral agreement with some colleges Limiting the number of students to be exchanged: Low participation rate

15/52 Outline Current Procedure Model Generalized Top Trading Cycles and Chains Minimum Quota

Model 16/52 A student exchange problem consists of a set of colleges C = {c 1,..., c m }, a set of students S = c C S c, an aggregate quota vector Q = (Q c ) c C, a list of student preferences P = (P s ) s S, a bilateral quota matrix q = (q c,d ) c,d C, q c,d :max number of college d students that college c would like to import an imbalance-tolerance vector b = (b c ) c C, b c : max level of dierence between incoming and outgoing students that college c can tolerate

Model 17/52 Denitions A (feasible) matching is a function µ : S C such that for any c, d C µ 1 (c) S d q c,d, µ 1 (c) Q c and µ 1 (c) µ 1 (C \ c) S c b c.

Model 17/52 Denitions A (feasible) matching is a function µ : S C such that for any c, d C µ 1 (c) S d q c,d, µ 1 (c) Q c and µ 1 (c) µ 1 (C \ c) S c b c. A matching µ is fair if there does not exist a student pair (s, s ) such that (1) {s, s } S c (2) s c s and (3) µ(s )P s µ(s). A matching is Pareto ecient if there does not exist another matching such that all students students are (weakly) better-o A mechanism is strategy-proof if students cannot benet from misreporting their preferences.

18/52 Outline Current Procedure Model Generalized Top Trading Cycles and Chains Minimum Quota

Generalized Top Trading Cycles and Chains 19/52 gttcc Top Trading Cycles and Chains introduced by Roth, Sonmez and Unver (2004) for kidney exchange A cycle is a list of colleges and students (c 1, s 1, c 2, s 2,..., c k, s k ) each agent points to the adjacent agent and the last student points to the rst college A chain is a list of colleges and students (c 1, s 1, c 2, s 2,..., s k 1, c k ) each agent points to the adjacent agent

Generalized Top Trading Cycles and Chains 20/52 gttcc Step 0: Assign a tie-breaking number to each college

Generalized Top Trading Cycles and Chains 20/52 gttcc Step 0: Assign a tie-breaking number to each college Step 1: Each college points to its own student with the highest internal priority

Generalized Top Trading Cycles and Chains 20/52 gttcc Step 0: Assign a tie-breaking number to each college Step 1: Each college points to its own student with the highest internal priority Each pointed student points to her favorite college which has available positions for students from her home college, available positions for all students, and imbalance below its tolerance level or remaining students. points to the students pointing itself. If there is a cycle: Execute each cycle. Remove assigned students. Update the quotas. Return to Step 1. Otherwise proceed to Step 2.

Generalized Top Trading Cycles and Chains 21/52 Properties of gttcc Theorem gttcc is fair, Pareto ecient and strategy-proof.

Generalized Top Trading Cycles and Chains 21/52 Properties of gttcc Theorem gttcc is fair, Pareto ecient and strategy-proof. Independent of the quotas set, each college has a balance no more than its tolerable level incoming students no more than its aggregate quota No need to set lower quotas = More exchanges compared to the current procedure

Generalized Top Trading Cycles and Chains 22/52 Simulations By using simulations, we compare current procedure and gttcc under various scenarios

Generalized Top Trading Cycles and Chains Simulations By using simulations, we compare current procedure and gttcc under various scenarios Fraction of exchanged students is more under gttcc More students prefer their assignment under gttcc 1 0.9 Exchange under Current Exchange under gttcc Preferring Current Preferring gttcc 0.8 0.7 0.6 Fraction 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Alpha α is the degree of correlation in preferences 22/52

23/52 Outline Current Procedure Model Generalized Top Trading Cycles and Chains Minimum Quota

Extensions 24/52 Minimum Quota We observe lack of diversity in exchanges

Extensions 24/52 Minimum Quota We observe lack of diversity in exchanges In order to increase the diversity (having incoming students from more countries) in the exchange we add minimum quota requirement to our model m c,d : is the minimum number of college d students to be assigned to college c.

Extensions 25/52 Adding one more step to gttcc We modify gttcc by adding an initial step in which before students point, we check whether there is enough students left from each college, there is enough available seats left in each college to satisfy all minimum quota requirement.

Extensions 26/52 gttcc with Minimum Quota Theorem gttcc with minimum quota is fair, strategy-proof and its outcome cannot be Pareto dominated by another matching satisfying minimum quota requirement.

Extensions Tie breaking:c-d-b-a a 2 a 1 b 2 b 3 A B a 3 b 1 c 1 d 3 C D c 3 c 2 d 1 d 2 a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 1 1 2 4 1 1 0 0 B 1 1 1 3 1 0 0 1 C 2 1 2 2 0 0 0 0 D 1 1 1 2 1 0 0 0 27/52

Extensions Tie breaking:c-d-b-a a 2 a 1 b 2 b 3 A B a 3 b 1 c 1 d 3 C D c 3 c 2 d 1 d 2 a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 1 2 2 4 1 1 0 0 B 1 1 1 3 1 0 0 1 C 2 1 2 2 0 0 0 0 D 1 1 1 2 1 0 0 0 28/52

Extensions Tie breaking:c-d-b-a a 2 b 2 b 3 A B a 3 d 3 C D c 3 c 2 d 2 a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 1 1 2 3 1 1 0 0 B 0 1 1 2 1 0 0 1 C 2 1 1 1 0 0 0 0 D 1 0 1 1 1 0 0 0 30/52

Extensions Tie breaking:c-d-b-a b 2 b 3 A B a 3 d 3 C D c 3 c 2 a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 1 1 1 2 1 1 0 0 B 0 1 1 2 1 0 0 1 C 2 1 1 1 0 0 0 0 D 0 0 1 0 1 0 0 0 33/52

Extensions Tie breaking:c-d-b-a b 2 b 3 A B d 3 C D c 3 c 2 a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 1 1 1 2 1 1 0 0 B 0 1 1 2 1 0 0 1 C 2 1 1 1 0 0 0 0 D 0 0 1 0 1 0 0 0 35/52

Extensions Tie breaking:c-d-b-a b 2 b 3 A B d 3 C D c 3 a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 1 1 1 2 1 1 0 0 B 0 1 1 2 1 0 0 1 C 2 1 1 1 0 0 0 0 D 0 0 1 0 1 0 0 0 37/52

Extensions Tie breaking:c-d-b-a b 3 A B d 3 C D a1 a2 a3 b1 b2 b3 c1 c2 c3 d1 d2 d3 B B B D A A A A D C A C C D D A B C B B B B B D C D A C q.,a q.,b q.,c q.,d Q b m..a m..b m..c m..d A 0 1 1 1 0 0 0 0 B 0 0 1 1 1 0 0 1 C 2 1 1 1 1 0 0 0 D 0 0 1 0 1 0 0 0 39/52

Extensions 41/52 Other Extensions gttcc mechanism can deal with many other constraints without losing its desirable properties Limit on the number of imports from a specic country/region Limit on the number of imports from a specic eld (engineering, management) Limit on the number of exports Tolerable Balance between incoming and outgoing students within each eld

Conclusion 42/52 Conclusion Consider a matching market which has not been studied to our knowledge Use TTCC mechanism in a many to one matching problem Fair (across to same types) Strategy-proof Ecient Welfare gains compared to the Current System without any cost

Conclusion 43/52 Back Up Slides Simulation Setup Example for Extensions gttcc with diversity

Conclusion 44/52 Simulations 10 Colleges and 100 Students. Each college has 10 students. Bilateral quotas: drawn from the discrete uniform distribution on the interval [0,4] Tolerance: drawn from the discrete uniform distribution on the interval [2,10] Aggregate quota: drawn from the discrete uniform distribution on the interval [Tolerance,10] Student preferences:u i,c = α Z c + (1 α) Z i,c Z c is an i.i.d standard uniformly distributed random variable and represents the common tastes of students on college c. Z i,c is also an i.i.d standard uniformly distributed random variable and represents the tastes of student i on college c. The correlation in the students preferences is captured by α [0, 1].

Conclusion 45/52 Simulations gttcc is calculated based on the true quotas and preferences In order to calculate the current system outcome we randomly create a bilateral quota vector, ( q c,c ) c C. for each college c such that qc,c = b c

Conclusion 46/52 Other Extensions b m 2 a e 1 A B a e 2 a m 3 C a e 1 a e 2 a m 3 b m 1 c m 1 B C C A A C B B q.,a q.,b q.,c Q b b e b m A 1 1 2 1 1 1 B 1 1 1 1 1 1 C 1 1 1 0 1 1 c m 1

Conclusion 47/52 Other Extensions b m 2 a e 1 A B a e 2 a m 3 C a e 1 a e 2 a m 3 b m 1 c m 1 B C C A A C B B q.,a q.,b q.,c Q b b e b m A 1 1 2 1 1 1 B 1 1 1 1 1 1 C 1 1 1 0 1 1 c m 1

Conclusion 48/52 Other Extensions b m 2 a e 1 A B a e 2 a m 3 C a e 1 a e 2 a m 3 b m 1 c m 1 B C C A A C B B q.,a q.,b q.,c Q b b e b m A 1 1 2 1 1 1 B 1 1 1 1 1 1 C 1 1 1 0 1 1 c m 1

Conclusion 49/52 Other Extensions A B a e 2 a m 3 C a e 1 a e 2 a m 3 b m 1 c m 1 B C C A A C B B q.,a q.,b q.,c Q b b e b m A 0 1 1 1 2 0 B 0 1 0 1 0 2 C 1 1 1 0 1 1 c m 1

Conclusion 50/52 Other Extensions A B a e 2 a m 3 C a e 1 a e 2 a m 3 b m 1 c m 1 B C C A A C B B q.,a q.,b q.,c Q b b e b m A 0 1 1 1 2 0 B 0 1 0 1 0 2 C 1 1 1 0 1 1 c m 1

Conclusion 51/52 Other Extensions A B a e 2 a m 3 C a e 1 a e 2 a m 3 b m 1 c m 1 B C C A A C B B q.,a q.,b q.,c Q b b e b m A 0 1 1 1 2 0 B 0 1 0 1 0 2 C 1 1 1 0 1 1 c m 1

Conclusion 52/52 Adding one more step to gttcc We modify gttcc by adding an initial step in which before students point, we check whether there is enough students left from each college, there is enough available seats left in each college to satisfy all minimum quota requirement. In particular, If # of remaining c students is equal to # of unlled reserved seats for c, then restrict all c students to point to the colleges with unlled reserved seats for c